rsx: tiny zcull optimizations

[scribam]

Should boost the performance on heavy zcull games.

[vietnam13231]

Infamous still needs to have "Disable ZCull Occlusion Queries" enabled for those to not overflow. So nothing here.

[Xcedf]

None of my zcull needing games react on this PR

[MSuih]

Burnout: Paradise has too agressive culling, which results in models popping in or appearing gray. Hard to show on picture, so here's a video: https://streamable.com/sr66x

Master and PS3 do not have any noticeable pop-in.

Edit: I'm getting same issues on Skate 3 as well, although not as extreme. I'm using RX 480, so perhaps this is an AMD-specific issue?

[kd-11]

Thank you @MSuih, I have also confirmed your findings. That helped figure out why there were performance gains; it was because zcull testing was illegally disabled.
@scribam
WRT the changes in check_zcull_status.
Picture the following:

1. The game changes the current depth buffer to some memory that is linked to the ZCULL unit. Zcull statistics are not enabled so no testing is expected. Your optimization aborts early and no ZCULL usage takes place.
2. The game enables Zcull testing. There is no framebuffer swapped however, so the set_active call now fails since zcull_surface_active was never updated and remains forever false.

EDIT: It is common for a game to enable and disable occlusion reports several times while rendering to the same framebuffer, even when it is bound to ZCULL ram. In fact, the framebuffer_swap path is very rarely taken in the context of a frame; usually about a dozen times at most for the highest end AAA games. Some rare cases can go above that but not by much since changing ROP target is quite expensive on GPU hardware (it messes with their memory controller caches)

[kd-11]

This breaks Zcull testing for reasons explained in comment.

[scribam]

Changes related to the Burnout regression reverted